Related Experiment Videos
Some neural network applications in environmental sciences. Part II: advancing computational efficiency of
Vladimir M Krasnopolsky1, Frédéric Chevallier
1Science Applications International Corporation at National Centers for Environmental Prediction, 5200 Auth Road, Camp Spring, MD 20746, USA. vladimir.kransnopolsky@noaa.gov
Summary
A novel neural network (NN) approach enhances computational efficiency and accuracy in environmental models. This method accelerates complex calculations for physical processes, offering a significant improvement for numerical modeling applications.
Area of Science:
- Environmental modeling
- Computational science
- Numerical methods
Background:
- Complex physical processes in environmental models require computationally intensive parameterizations.
- Existing parameterization algorithms often involve intricate mathematical expressions and nonlinear relations.
- Accelerating these calculations is crucial for improving the efficiency and accuracy of numerical simulations.
Purpose of the Study:
- To introduce a generic neural network (NN) application for improving computational efficiency in numerical environmental models.
- To demonstrate the NN approach's ability to approximate complex parameterizations and provide Jacobians.
- To present real-world applications of NNs in oceanic, atmospheric, and wave models.
Main Methods:
- Utilizing neural networks (NNs) as continuous mappings to replace primary parameterization algorithms.
- Applying NNs to approximate the UNESCO equation of state for seawater density and its inversion for salinity.
- Implementing NN approximations for longwave radiative transfer in atmospheric models and nonlinear wave-wave interactions in wave models.
Main Results:
- Significant acceleration of numerical computations across all four presented applications.
- NNs provided fast and accurate approximations of primary parameterizations.
- NNs efficiently generated Jacobians with minimal computational cost.
Conclusions:
- The NN approach offers a numerically efficient solution for frequently repeated, complex calculations in environmental models.
- This method can be broadly applied to physical, chemical, and biological processes within numerical models.
- NNs enhance both the speed and accuracy of environmental simulations, with demonstrated success in operational systems.